Stochastic binary synapses having sigmoidal cumulative distribution functions for unsupervised learning with spike timing-dependent plasticity.

Spike timing-dependent plasticity (STDP), which is widely studied as a fundamental synaptic update rule for neuromorphic hardware, requires precise control of continuous weights. From the viewpoint of hardware implementation, a simplified update rule is desirable. Although simplified STDP with stochastic binary synapses was proposed previously, we find that it leads to degradation of memory maintenance during learning, which is unfavourable for unsupervised online learning. In this work, we propose a stochastic binary synaptic model where the cumulative probability of the weight change evolves in a sigmoidal fashion with potentiation or depression trials, which can be implemented using a pair of switching devices consisting of serially connected multiple binary memristors. As a benchmark test we perform simulations of unsupervised learning of MNIST images with a two-layer network and show that simplified STDP in combination with this model can outperform conventional rules with continuous weights not only in memory maintenance but also in recognition accuracy. Our method achieves 97.3% in recognition accuracy, which is higher than that reported with standard STDP in the same framework. We also show that the high performance of our learning rule is robust against device-to-device variability of the memristor's probabilistic behaviour.

Magnetization switching of a perpendicular nanomagnet induced by vertical nonlocal injection of pure spin current.

Injection of pure spin current using a nonlocal geometry is a promising method for controlling magnetization in spintronic devices from the viewpoints of increasing freedom in device structure and avoiding problems related to charge current. Here, we report an experimental demonstration of magnetization switching of a perpendicular magnetic nanodot induced by vertical injection of pure spin current from a spin polarizer with perpendicular magnetization. In comparison with direct spin injection, the current amplitude required for magnetization switching is of the same order and shows smaller asymmetry between parallel-to-antiparallel and antiparallel-to-parallel switching. Simulation of spin accumulation reveals that, in the case of nonlocal spin injection, the spin torque is symmetric between the parallel and antiparallel configuration because current flows through only the spin polarizer, not the magnetic nanodot. This characteristic of nonlocal spin injection is the origin of the smaller asymmetry of the switching current and can be advantageous in spintronic applications.

Zero-dc-field rotation-direction-dependent magnetization switching induced by a circularly polarized microwave magnetic field.

Magnetization switching of high-anisotropy nanomagnets by a small magnetic field is a key challenge in developing future magnetic nanodevices. In this paper, we experimentally demonstrate magnetization switching of a perpendicularly magnetized nanomagnet induced solely by an in-plane circularly polarized microwave magnetic field. Applying a microwave field with an amplitude below 5% of the anisotropy field induces large ferromagnetic resonance excitation, which results in magnetization switching even in the absence of a dc field. This kind of magnetization switching is induced by a microwave field with a duration of 0.5 ns and is clearly dependent on the rotation direction of the microwave field.

Nanoscale layer-selective readout of magnetization direction from a magnetic multilayer using a spin-torque oscillator.

Technology for detecting the magnetization direction of nanoscale magnetic material is crucial for realizing high-density magnetic recording devices. Conventionally, a magnetoresistive device is used that changes its resistivity in accordance with the direction of the stray field from an objective magnet. However, when several magnets are near such a device, the superposition of stray fields from all the magnets acts on the sensor, preventing selective recognition of their individual magnetization directions. Here we introduce a novel readout method for detecting the magnetization direction of a nanoscale magnet by use of a spin-torque oscillator (STO). The principles behind this method are dynamic dipolar coupling between an STO and a nanoscale magnet, and detection of ferromagnetic resonance (FMR) of this coupled system from the STO signal. Because the STO couples with a specific magnet by tuning the STO oscillation frequency to match its FMR frequency, this readout method can selectively determine the magnetization direction of the magnet.